Search

Hyperspectral pansharpening (HP) can fuse different kinds of remote sensing images for the downstream classification or detection task. Deep learning methods have made rapid development in HP. However, most deep learning-based HP methods design the network architecture without considering the specific downstream task. When the pansharpening image is transferred to the downstream tasks, it may lead to poor performance on the downstream tasks. To solve this problem, a novel classification-oriented multi-task network (CMNet) is proposed for HP. CMNet combines the classification and HP networks in a multi-task learning way. In the classification network, class-specific and spatial-difference contrastive constraints are defined to enforce the extraction module to extract discriminative features for classification. In the HP network, a fusion module with an identical structure to the extraction module is constructed to extract fused features for reconstruction. Then, multi-gradient learning is designed to optimize the HP network under the guidance of the classification network. It adopts gradient mapping to consider the conflicting or collaborative situations between these different tasks. In this way, the fusion module in the HP network learns discriminative feature extraction ability from the classification network, which makes the HP network realize classification-oriented HP. Experimental results demonstrate that the proposed method markedly improves the pansharpening performance and the pansharpening is also able to elevate the classification performance. [ABSTRACT FROM AUTHOR]

Considering that saliency areas in images play important roles in human visual perception, in this paper we propose a Learning Compressive Sampling (LCS) scheme and its physical implementation for more efficient image acquisition via a new Multiscale and Steerable Support Value Transform (MS2VT). The key idea is to learn geometric saliency maps of images by MS2VT, which is deduced from a mapped least square-support vector machine. Because MS2VT can produce a multiscale, multidirectional, undecimated, dyadic and aliasing transform with shift-invariant and anisotropy properties, the obtained support values can reveal the geometric and saliency information of images. The learned saliency map is then used to formulate a variable density compressive function, to realize a simple, fast and efficient sampling, which aims to allocate more sensing resources to saliency attention areas but fewer to non-salient regions. Several experiments are taken on some natural images and remote sensing images to compare our proposed LCS scheme with traditional samplings when saliency information is not used. Moreover, the performance of MS2VT based saliency detection scheme is also compared with other related saliency detection approaches. The experimental results indicate that it can obtain high quality images, especially in preserving more detailed edges, contours and complex structures, even at low sampling ratios. [ABSTRACT FROM AUTHOR]